---
license: mit
tags:
- reinforcement-learning
- ppo
- gymnasium
- deep-rl-course
- LunarLander-v2
model-index:
- name: PPO LunarLander-v3
  results:
  - task:
      type: reinforcement-learning
      name: Reinforcement Learning
    dataset:
      name: LunarLander-v3
      type: LunarLander-v3
    metrics:
    - type: mean_reward
      value: -229.885
      name: Mean Reward
---
# PPO from scratch — LunarLander-v3

Trained using a CleanRL-style PyTorch PPO implementation adapted for Gymnasium.

Evaluation over 10 episodes: **-229.89 ± 22.10**.

- `model.pt`: actor-critic PyTorch state dict
- `ppo.py`: full training/evaluation/upload source
- `replay.mp4`: evaluation video (if rendering succeeded)
- `hyperparameters.json`: training settings
- `eval_results.json`: evaluation results

To run locally, install PyTorch, `gymnasium[box2d]`, `imageio`, `imageio-ffmpeg`, `tensorboard`, and `huggingface_hub`.